Lewis B. Sheiner, MD
Depts of Laboratory Medicine, Medicine & Pharmacy UCSF, San Francisco, California, and Rhone-Poulenc-Rorer Research and Development, Antony, France
Modeling, in the context of drug development, means the application of theory (science) as opposed to, or in addition to, strict empiricism. When one assumes that a smooth asymptotic functional relationship exists between a pharmacologic effect and drug exposure, and exploits this assumption to design and analyse clinical trials, one is using (PK/PD) theory. If, in contrast, one assumes that each different exposure may produce any possible degree of response, regardless of the response exhibited by similar exposures, and one designs and analyses a clinical trial under this view, one is using strict empiricism. Reliable theory allows parsimonious and flexible experimental designs and efficient estimation (inference), at the price of making assumptions that are not (often cannot be) tested on the data at hand. If the assumptions are false, so may be the conclusions. In contrast, strict empiricism offers greater assurance by making only few and unquestioned assumptions, at the price of rigid designs and large data requirements. The question, then, is what should be the balance between theory and empiricism in the clinical evaluation stages of the drug development process?
Consider the essential questions that must be answered in humans before a drug can finally be accepted as useful:
- Does the drug appear to work (i.e., is short-term efficacy short term toxicity)? If the answer to this question is affirmative, then
- How does one individualize therapy to attain good individual short-term benefit/risk (i.e., what is the response surface, the relationships among dosage, patient-specific characteristics, and short-term efficacy and toxicity)? Finally,
- Under conditions of usual use, does net (societal) benefit exceed harm (i.e., is there ultimate efficacy)?
The first and last questions may be addressed efficiently using theory, but there is a good argument for being largely empirical instead: the questions are highly focussed, and the answers have great import. Regarding the first question, one does not wish to embark upon a costly development scheme (which is the result of an affirmative answer to this question) without strong evidence of promise; evidence that does not depend on questionable assumptions. Nor is any theory of clinical action of drugs so sure that we may omit an empirical test of a therapy once it has begun to be used widely (the recent CAST results provide an unfortunate example), although it can be argued that such trails should be for drug classes, not individual drugs, post-approval, and at public expense (as in the CAST example).
Once, however, the first question has been answered in the affirmative, we must learn how to use the drug well with respect to surrogate endpoints and relatively rapid appearing toxicities, and begin to do so before we can address the third question. The second question, then, requires not skepticism, but rather the use of all available knowledge and data if one is to be successful without expending huge resources. The reason is this: dosage choice involves controlling several factors (amount, sechedule) while patient variability is also multi-factorial (sex, age,other drugs, etc). One cannot practically define the response surface (even if outcome were univariate, which it is not), empirically: the combinatorics guarantee that the demand for data quickly becomes overwhelming. Thus, model (theory) based design and analysis of drug trials for the purpose of determining how to individualize dosage is not only sensible; it is, in fact, necessary.
Recognizing, however, the continuing requirement that the drug evaluation process lead to timely approval by regulatory authorites, and the value of assumption-free inference in such a process, one is led to consider designs for human trials that permit both sharp empirical inferences and, at the same time, exploration of the response surface. Use of more informative designs of this kind, plus increased flexibility on the part of regulatory authorities, who must learn to balance the need for empirical certainty against the power of theory-based approaches, will encourage the development and application of useful models so that drug development can become more rational, efficient, and informative.
A reasonable pursuit for PAGE participants, then, is to explore and encourage study designs oriented towards the transition from efficacy trials to response-surface elucidation, thereby providing the real-world experience with such designs and the population-based PK/PD modeling their analysis entails on which further evolution of clinical drug development programs can be based.
Reference: PAGE 3 () Abstr 879 [www.page-meeting.org/?abstract=879]
Poster: oral presentation